| In recent years,the postgraduate study has become a hot choice for students Postgraduate education is developing rapidly and the number of supervisors and students is gradually huge.However,there is a general lack of understanding of the comprehensive quality and cultivation style of supervisors among graduate students,which makes them choose the wrong supervisors.Coupled with the fail of good communication between supervisors and graduate students,the problem of disharmony between supervisors and students is more serious,which makes the cultivation work more difficultFor this reason,based on the concept of educational evaluation,this thesis analyzes the current status of supervisor evaluation and supervisor-student relationship based on the analysis of the needs of postgraduate students and supervisors.From the subjective perspective of postgraduate students,a supervisor scoring model based on web-based anonymous evaluation is constructed.And based on the supervisor scoring model,we further design and develop the supervisor scoring system based on web-based anonymous evaluation,and present the evaluation results in the system after visualizationIn terms of the rating model research problem,this study combines the features of the review text,extracts the high-frequency words from the postgraduate reviews,and summarizes the six evaluation dimensions.At the same time,natural language processing methods and deep learning techniques are introduced,integrating the word2vec model and LSTM model to semantic analysis and sentiment classification training of unstructured review texts,to obtain the scores of each dimension of supervisors.Combining the scores of each dimension,the entropy weight method is used to calculate the weights of each dimension,and finally,the overall score of the supervisor is calculatedThe scoring model is applied by selecting the comment contents of the supervisor evaluation website and setting up several comparison experiments,to test the validity of the supervisor scoring model.For model training as well as model testing,five postgraduates are invited to manually annotate some of the comment data and to test the reliability of the annotators.After the analysis of the annotation results,Kendall’s concordance coefficient(W)is 0.873,and the reliability,as well as the consistency of the annotation results,are strong.And then,the manually annotated data are used as a test of the classification results,and different word vector training methods and different sentiment classification methods are used for training.The results of the comparison experiments show that compared with different word vector training methods,the method used in this research is better than the traditional word vector method and n-gram method in terms of accuracy and recall,with 17.13%and 11.66%higher accuracy,respectively.Compared with different sentiment classification methods,the macro-F1 value of the method used in this study is 90.53%,which is better than the traditional methods of CNN,RNN,and TextCNN in terms of the performance of macro-F1 value.It is initially shown that the supervisor scoring model proposed in this thesis is effective in mining the sentiment scores of the review texts of postgraduates On this basis,this thesis combines the constructed supervisor scoring model with web crawlers,web development,data visualization,and other technologies to design and develop a supervisor scoring system based on a web-based anonymous evaluation to display supervisor evaluation information in all aspects.It provides some help to postgraduates and their supervisors in the issue of supervisor-student relationshipsThis thesis has attempted to apply web-based anonymous evaluation to supervisor evaluation,and this attempt can help to improve the problem of disharmony in the supervisor-student relationship due to the lack of authentic understanding and communication between supervisors and students to some extent.Future research will collect a larger amount of data with more dimensions and bring in more factors for further exploration of the study. |
